Attribute Analysis and Diagnosis of LUNG CT Images of COVID-19

Abstract

Computed Tomography (CT) is an authoritative verification standard for patients with Corona Virus Disease 2019 (COVID-19). Automatic detection of lung infection through CT is of great significance for epidemic prevention and control and prevention of cross-infection. The accuracy of existing lung CT image segmentation methods is not high, and due to the privacy protection measures of hospitals, the number of COVID-19 lung CT data sets is too small, which is prone to over-fitting during training. In this paper, we propose a qualitative mapping model for the diagnosis and localization of COVID-19 lesions. The binary image processed by U-net network is used as input, and lung CT is segmented as four attributes, and attribute diagnosis is carried out with the help of correlation matrix and transformation degree function. Experiments show that this method not only avoids the over-fitting risk of data sets, but also increases the robustness of data. Experiments also prove that this design has higher accuracy than the simple neural network learning.